Design And Development Of Bank Card Recognition System Based On Deep Learning

In recent years,with the rapid development of China's science and technology and the advent of the internet age,mobile payment and various mobile banking services have become increasingly popular.In this field,Alipay and WeChat already have applications of using mobile phone photos to identify bank card.More and more businesses are also trying to apply this technology.The bank card identification system designed in this paper includes four modules:bank card number line positioning,card number line preprocessing and character segmentation,deep learning based character recognition and GUI interface design.First of all,in this paper,the card number line database,the bank card character database and the bank card encoding rules database are established.In the positioning stage of the card number line,the sample feature of the card number line is extracted by combining the HOG operator and the SVM classifier.For the possible multi-region positioning problem,the non-maximum suppression algorithm is used to eliminate the mis-location area,and the positioning function of the bank card number line is realized.In the card number preprocessing and character segmentation stage,the bank card is divided into two types:black body printing and embossing printing.The conventional image processing technology is used to eliminate the interference information for the black body printing,and it is found that the method of pixel point conversion can more completely remove the background interference during the experiment.The card number line of the embossed printing type is separated by the color space to separate the background pattern.Then,the Canny edge detection is used to extract the character information.Finally,the card numbers of the two printing types are vertically projected to separate character.In the character recognition phase,the mainstream TensorFlow deep learning framework and the LeNet-5 deep convolutional neural network are used.Firstly,the TensorFlow experimental platform was built,and then the original LeNet-5 network was used for character recognition.However,the experiment found that the network has low recognition rate for the character.So,the LeNet-5 neural network structure is improved,the network depth is increased,and the activation function is modified.After the character recognition stage is completed,according to the collected bank card coding rules database to judge the card type and card issuing unit.Finally,based on PYQT5 to write GUI interface and display functions.